**** BEGIN LOGGING AT Wed Jun 27 03:00:01 2018
Jun 27 13:15:11 <greguu> ant_work: hi
Jun 27 13:18:34 <ant_work> hello
Jun 27 13:20:05 <greguu> have you had a chance to try 1.1.19?
Jun 27 13:20:09 <greguu> musl
Jun 27 13:20:44 <ant_work> yes, miscompiled, alignment traps
Jun 27 13:21:34 <greguu> hm, I fear the issue is somewhere else
Jun 27 13:21:50 <greguu> cflags or what gcc
Jun 27 13:22:06 <ant_work> khem thinks that one commit in musl did expose a gcc bug
Jun 27 13:22:27 <greguu> but it does not affect my builds, all with 1.1.19 and gcc 7.3.0
Jun 27 13:22:34 <ant_work> I have to write a testcase, he'll submit upstream
Jun 27 13:23:00 <ant_work> on our toolchain, bug appears with both gcc7.3 and 8.1
Jun 27 13:23:00 <greguu> aligment issues point to ABI ?
Jun 27 13:23:50 <greguu> is musl patched on OE ?
Jun 27 13:23:55 <greguu> or gcc _
Jun 27 13:24:04 <ant_work> gcc is
Jun 27 13:25:05 <greguu> could you try other cflags ?
Jun 27 13:40:28 <ant_work> I have tried without luck. I'll send to you the code snippet to test as first
Jun 27 13:40:47 <ant_work> hope to do that this evening
Jun 27 13:42:22 <ant_work> here the patches:
Jun 27 13:42:23 <ant_work> http://cgit.openembedded.org/openembedded-core/tree/meta/recipes-devtools/gcc
Jun 27 13:44:00 <ant_work> if you have binutils-cross around, try to objdum -d -S libc.so > musl.dump
Jun 27 13:44:14 <ant_work> then search for   readdir
Jun 27 13:45:14 <ant_work> you'll see smthg like
Jun 27 13:45:16 <ant_work> 0001d028 <readdir>:
Jun 27 13:45:16 <ant_work>    1d028:	e5902010 	ldr	r2, [r0, #16]
Jun 27 13:45:16 <ant_work>    1d02c:	e5901014 	ldr	r1, [r0, #20]
Jun 27 13:45:16 <ant_work>    1d030:	e280c020 	add	ip, r0, #32
Jun 27 13:45:27 <ant_work> that last one is the key,   #32
Jun 27 13:45:41 <ant_work> after that musl-commit it becomes #28
Jun 27 13:46:05 <ant_work> so th efurther instruction gets unaligned address
Jun 27 13:46:06 <ant_work> 1d070:	e1cc00d8 	ldrd	r0, [ip, #8]
Jun 27 13:46:52 <ant_work> For STRD and LDRD, the specified address must be word-aligned
Jun 27 13:47:07 <ant_work> beats me how that happens...
Jun 27 13:47:35 <ant_work> this C
Jun 27 13:47:36 <ant_work> https://git.musl-libc.org/cgit/musl/tree/src/dirent/readdir.c
Jun 27 13:47:56 <ant_work> I have to extract a testcas of it
Jun 27 13:50:33 <ant_work> basically you hit it with many commands...top, ls, ...
Jun 27 13:50:36 <ant_work> pid
Jun 27 13:55:14 <greguu> hm
Jun 27 13:55:18 <greguu> 000000000001faa0 <readdir>:
Jun 27 13:55:20 <greguu>    1faae: 7d 20                 jge    1fad0 <readdir+0x30>
Jun 27 13:55:22 <greguu>    1fae4: 7e 0a                 jle    1faf0 <readdir+0x50>
Jun 27 13:55:24 <greguu>    1faec: eb c8                 jmp    1fab6 <readdir+0x16>
Jun 27 13:55:26 <greguu>    1faf2: 79 1c                 jns    1fb10 <readdir+0x70>
Jun 27 13:55:28 <greguu>    1faf7: 74 17                 je     1fb10 <readdir+0x70>
Jun 27 13:55:30 <greguu>    1fb04: eb c2                 jmp    1fac8 <readdir+0x28>
Jun 27 13:55:32 <greguu> so mine is 28
Jun 27 13:55:34 <greguu> but all works
Jun 27 13:56:47 <greguu> looks different to your dump though
Jun 27 13:58:57 <ant_work> I've sent now good & bad dumps
Jun 27 13:59:14 <greguu> I need to do it native ?
Jun 27 13:59:24 <ant_work> no, cross
Jun 27 14:00:04 <ant_work> people at gcc do not want binaries, just small code snippets...
Jun 27 14:04:46 <ant_work> that above is x86, you should have an arm-xx-linux-musleabi-objdump installed by binutils-cross
Jun 27 14:07:12 <greguu> hm, still very different
Jun 27 14:07:15 <greguu> 0001f924 <readdir>:
Jun 27 14:07:17 <greguu>    1f93c: ba000012  blt 1f98c <readdir+0x68>
Jun 27 14:07:19 <greguu>    1f964: ca000008  bgt 1f98c <readdir+0x68>
Jun 27 14:07:36 <greguu> that was with voidz-packages/masterdir-armv5tel-musl/bin/arm-linux-musleabi-objdump -d -S libc.so > musl.dump
Jun 27 14:09:47 <ant_work> is that libc.so the one for arm?
Jun 27 14:09:56 <greguu> yes
Jun 27 14:10:14 <greguu> I do it native now, but it takes a looong time
Jun 27 14:10:25 <ant_work> the issue is only with ldrd/strd
Jun 27 14:11:35 <greguu> hm, does not appear in my libc.so
Jun 27 14:11:39 <greguu> maybe cflags
Jun 27 14:13:26 <greguu> my native
Jun 27 14:13:28 <greguu> 0001f924 <readdir>:
Jun 27 14:13:30 <greguu>    1f93c: ba000012  blt 1f98c <readdir+0x68>
Jun 27 14:13:32 <greguu>    1f964: ca000008  bgt 1f98c <readdir+0x68>
Jun 27 14:13:34 <greguu> same
Jun 27 14:13:45 <greguu> so toolchain / cflags / gcc ?
Jun 27 14:13:46 <greguu> not musl
Jun 27 14:14:47 <greguu> armv5 thing
Jun 27 14:14:49 <greguu> or so
Jun 27 14:15:10 * greguu is gone, its late here
Jun 27 14:29:32 <ant_work> it is the 'dsp' tune
Jun 27 14:29:43 <ant_work> the e in armv5te
Jun 27 14:30:15 <ant_work> these ldrd/strd are dsp instructions only present on armv5_e_
Jun 27 14:30:28 <ant_work> so you are not optimizing for dsp...
Jun 27 14:31:47 <ant_work> check pls your   -march=armv5te
Jun 27 14:54:23 <ant_work> bbl
**** ENDING LOGGING AT Thu Jun 28 03:00:02 2018